Mapping string/unicode dtypes in numpy arrays about reticulate HOT 10 CLOSED

I think string and unicode would be R character vectors, and object would be a character vector if all of the elements are strings (otherwise just a list of objects)

from reticulate.

Is the data type you are looking for here basically a NumPy string array?

from reticulate.

@jjallaire correct.

from reticulate.

Okay, in your experience do pandas data frames often use dtype=object (for variable length strings) or do they more typically stick to a fixed string size?

from reticulate.

Unfortunately I don't work with Python enough to opine on that, although from browsing SO it seems to often be the case. Are we thinking about upcasting all string types to object or finding some reasonable superset of string types?

from reticulate.

Sounds good!

from reticulate.

This should be resolved by this commit: 3868a8d

from reticulate.

@jjallaire this crashes:

np <- import("numpy")
np$array(c("a", "b"), dtype = "S256")

from reticulate.

Stack trace:

* thread #1: tid = 31749, 0x00007ffff07d3419 libpython3.6m.so`PyBytes_AsStringAndSize(obj=0x0000000000000000, s=0x00007fffffffacd0, len=0x00007fffffffacd8) + 9 at bytesobject.c:1289, name = 'R', stop reason = signal SIGSEGV: invalid address (fault address: 0x8)
    frame #0: 0x00007ffff07d3419 libpython3.6m.so`PyBytes_AsStringAndSize(obj=0x0000000000000000, s=0x00007fffffffacd0, len=0x00007fffffffacd8) + 9 at bytesobject.c:1289

from reticulate.

Should be fixed now, thanks for the report!

from reticulate.